Fast and accurate algorithm for eye localisation for gaze tracking in low‐resolution images

Abstract

Iris centre localization in low-resolution visible images is a challenging problem in computer vision community due to noise, shadows, occlusions, pose variations, eye blinks, etc. This paper proposes an efficient method for determining iris centre in low-resolution images in the visible spectrum. Even low-cost consumer-grade webcams can be used for gaze tracking without any additional hardware. A two-stage algorithm is proposed for iris centre localization. The proposed method uses geometrical characteristics of the eye. In the first stage, a fast convolution based approach is used for obtaining the coarse location of iris centre (IC). The IC location is further refined in the second stage using boundary tracing and ellipse fitting. The algorithm has been evaluated in public databases like BioID, Gi4E and is found to outperform the state of the art methods.

Detecting Gaze Direction Using Robot-Mounted and Mobile-Device Cameras

“…Additionally, algorithms for remote eye-trackers need to deal with off-axis camera positions or partly visible eyes. From the perspective of our research, three algorithms for remote eye-tracking are of special interest: one proposed by George and Routray, another proposed by Droege and Paulus, and a third by Timm and Barth [3,7,13]. Among these three algorithms, the approach by Timm and Barth achieved the best results, achieving stable detection rates on all three data sets [4].…”

“…There are many articles describing the approaches to gaze tracking for a robot, but most of them do not provide any working samples that can be evaluated [7,8,17,18]. Some commercial solutions are available, but they are based on commercial eye trackers like Tobii or Pupil Labs.…”

Convolutional Neural Network Implementation for Eye-Gaze Estimation on Low-Quality Consumer Imaging Systems

“…It was trained with 2000 image/gaze position pairs. A hybrid approach is adopted in [35] in which the iris centres are determined first using circular a Hough transform, followed by refinement using a gradient-aware random sample consensus (RANSAC) algorithm and ellipse fitting. Eye corners are estimated using Gabor jets [36] and tracked using optical flow with normalized cross-correlation.…”

Hybrid eye center localization using cascaded regression and hand-crafted model fitting

“…Subsequently, an ellipse is fit to the pupil using RANSAC. Wood and Bulling [25], as well as George and Routray [11], have a similar scheme but employ a voting-based approach to get an initial eye center estimate. Fuhl et alpropose the Excuse [9] and Else [10] algorithms.…”

“…The first, predominant, category consists of hand-crafted model fitting methods. These techniques employ the appearance, such as the darkness of the pupil, and/or the circular shape of the pupil and the iris for detection [3,9,10,11,16,19,20,22,24,25]. These methods are typically accurate but often lack robustness in more challenging settings, such as low resolution or noisy images and poor illumination.…”